Semiautomated Segmentation of Kidney From High-Resolution Multidetector Computed Tomography Images Using a Graph-Cuts Technique

Abstract

To develop a semiautomated segmentation method based on a graph-cuts technique from multidetector computed tomography images for kidney segmentation and to evaluate and compare it with the conventional manual delineation segmentation method. We have developed a semiautomated segmentation method that is based on a graph-cuts technique with enhanced features including automated seed growing. Multidetector computed tomography images were obtained from 15 consecutive patients who were being evaluated as possible living donors for kidney transplant. Two observers independently performed the segmentation of the kidney from the multidetector computed tomography images using the manual and semiautomated methods. The efficiency of the 2 methods were measured by segmentation processing times and then compared. The interobserver and method reproducibility was determined by Dice similarity coefficient (DSC), which measures how closely 2 segmented volumes overlap geometrically and the coefficient of variation of volume measurements. The mean segmentation processing time was (manual vs semiautomated, P < 0.001) 96.8 +/- 13.6 vs 13.7 +/- 3.5 minutes for observer 1 and 44.3 +/- 4.7 vs 16.2 +/- 5.1 minutes for observer 2. The mean interobserver reproducibility was (manual vs semiautomated, P < 0.001) 93.6 +/- 1.6% vs 97.3 +/- 0.9% for DSC and 5.3 +/- 2.6% vs 2.2 +/- 1.3% for coefficient of variation, indicating higher interobserver reproducibility with the semiautomated than manual method. The agreement between the 2 segmentation methods was high (mean intermethod DSC 95.8 +/- 1.0% and 94.9 +/- 0.8%) for both observers. The semiautomated method was significantly more efficient and reproducible than the manual delineation method for segmentation of kidney from MDCT images.